Through a prospective clinical sequencing program for advanced cancers, four index cases were identified which harbor gene rearrangements of FGFR2, including patients with cholangiocarcinoma, breast cancer, and prostate cancer. After extending our assessment of FGFR rearrangements across multiple tumor cohorts, we identified additional FGFR fusions with intact kinase domains in lung squamous cell cancer, bladder cancer, thyroid cancer, oral cancer, glioblastoma, and head and neck squamous cell cancer. All FGFR fusion partners tested exhibit oligomerization capability, suggesting a shared mode of kinase activation. Overexpression of FGFR fusion proteins induced cell proliferation. Two bladder cancer cell lines that harbor FGFR3 fusion proteins exhibited enhanced susceptibility to pharmacologic inhibition in vitro and in vivo. Because of the combinatorial possibilities of FGFR family fusion to a variety of oligomerization partners, clinical sequencing efforts, which incorporate transcriptome analysis for gene fusions, are poised to identify rare, targetable FGFR fusions across diverse cancer types.
Significance: High-throughput sequencing technologies facilitate defining the mutational landscape of human cancers, which will lead to more precise treatment of patients with cancer. Here, through integrative sequencing efforts, we identified a variety of FGFR gene fusions in a spectrum of human cancers. FGFR fusions are active kinases. Cells harboring FGFR fusions showed enhanced sensitivity to the FGFR inhibitors PD173074 and pazopanib, suggesting that patients with cancer with FGFR fusions may benefit from targeted FGFR kinase inhibition. Cancer Discov; 3(6); 636–47. ©2013 AACR.
See related commentary by Sabnis and Bivona, p. 607
This article is highlighted in the In This Issue feature, p. 591
Advances in next-generation sequencing technologies have refined the molecular taxonomy of a spectrum of human diseases and facilitated a move toward “precision medicine” (1, 2). With regard to oncology, defining the mutational landscape of an individual patient's tumor will lead to more precise treatment and management of patients with cancer. Comprehensive clinical sequencing programs for patients with cancer have been initiated at a variety of medical centers, including our own (3, 4). In addition to the potential for identifying “actionable” therapeutic targets in patients with cancer, these clinical sequencing efforts may lead to the identification of novel “driver” mutations that may be rare in a common cancer type or newly revealed in relatively rare cancer types.
Recurrent gene fusions are an important class of “driver” mutation in cancer, as exemplified by the BCR–ABL gene fusion that characterizes chronic myeloid leukemia (CML; ref. 5). Importantly, virtually all patients with CML harbor the BCR–ABL kinase fusion and respond to the small-molecule kinase inhibitor, imatinib, representing one of the earliest examples of precision medicine in practice (6). In 2005, it was discovered that more than 50% of prostate cancers harbor recurrent gene fusions of the androgen-regulated gene TMPRSS2 with ETS transcription factors (7), suggesting that gene fusions/translocations may play a significant role in common epithelial tumors, similar to hematologic malignancies and sarcomas. Subsequently, recurrent gene rearrangements have been identified in carcinomas of the lung, breast, colon, and thyroid, among other epithelial tissues (8–12). Of these, the EML4–ALK gene fusion, which characterizes 1% to 5% of lung adenocarcinomas, has gained the most traction in the context of precision therapy, as patients with this gene fusion respond to the kinase inhibitor crizotinib (13, 14). Recently, FGFR1 and FGFR3 fusions with TACC1 and TACC3, respectively, have been identified in approximately 3% of the tumor glioblastoma multiforme (GBM; ref. 15), and FGFR3–TACC3 fusions were identified in a subset of bladder carcinomas (16). Preclinical studies suggest that patients with GBM with FGFR–TACC gene fusions may benefit from targeted FGFR kinase inhibition (17, 18).
Our Institutional Review Board (IRB)-approved clinical sequencing program, called the Michigan Oncology Sequencing Program (MI-ONCOSEQ), enrolls patients with advanced cancer across all histologies (3). Since April 2011, we have enrolled more than 100 patients on this program, which involves obtaining a current tumor biopsy with matched normal samples (blood and/or buccal swab). The samples are then subjected to integrative sequencing that includes whole-exome sequencing of the tumor and matched normal tissue, transcriptome sequencing, and, as needed, low pass genome sequencing (3). This combination of DNA and RNA sequencing technologies allows one to be relatively comprehensive with regard to the mutational landscape of coding genes, including point mutations, indels, amplifications, deletions, gene fusions/translocations, and outlier gene expression. These results are generated within a 5 to 7 week time frame and are presented at an institutional “precision tumor board” (previously called sequencing tumor board) to deliberate upon potentially actionable findings.
In this study, 4 MI-ONCOSEQ patients who harbored gene fusions of FGFR2 by transcriptome sequencing were prospectively identified (Fig. 1). The first patient (MO_1036) was a 34-year-old female diagnosed with metastatic cholangiocarcinoma. By whole-exome sequencing of the tumor relative to the matched normal, we detected 8 nonsynonymous somatic point mutations (Supplementary Table S1). The most interesting of these in terms of tumor biology was the inactivation of the SWI/SNF chromatin remodeling complex through mutation of ARID1A (Q1573*) and PBRM1 (C736*). The SWI/SNF complex has been implicated as a tumor suppressor, and inactivating somatic mutations of ARID1A and PBRM1 have been identified in renal cell carcinoma, breast cancer, and ovarian cancer (19). The copy number landscape for MO_1036 as determined by whole-exome sequencing is shown in Fig. 1A and Supplementary Table S2. Interestingly, by paired-end RNA sequencing, we detected an intrachromosomal fusion that resulted in the in-frame fusion of the FGFR2 to BICC1 (Fig. 1A). Although 7 additional chimeric RNAs were detected (Supplementary Table S3), only the FGFR2–BICC1 fusion exhibited a combination of high supporting reads (n = 259), predicted in-frame fusion protein, and predicted potential therapeutic actionability via kinase inhibition. The FGFR2–BICC1 fusion was confirmed by quantitative PCR (qPCR) analysis (Fig. 1A). Neither copy number aberrations nor point mutations were observed in FGFR2 or BICC1.
The second MI-ONCOSEQ patient with an FGFR2 fusion (MO_1039) was a 61-year-old male with metastatic cholangiocarcinoma. Like the first patient, this individual's tumor expressed an FGFR2–BICC1 fusion of identical configuration (Fig. 1B and Supplementary Table S4). This fusion was similarly validated by qPCR (Fig. 1B). In contrast, however, this cholangiocarcinoma case exhibited 27 nonsynonymous somatic point mutations, including an inactivating mutation of TP53 (R267W; Supplementary Table S5) and a distinct copy number landscape (Fig. 1B and Supplementary Table S6). Neither point mutations nor copy number changes in FGFR genes were identified in this patient.
The third patient with an FGFR2 fusion was a 31-year-old woman with metastatic breast cancer (MO_1051). RNA sequencing revealed an in-frame interchromosomal fusion of FGFR2 with AFF3, which had a functional structure analogous to the FGFR2 kinase fusions found in cholangiocarcinoma (Fig. 1C). In addition to the FGFR2–AFF3 fusion, which was detected with 138 supporting reads and validated by qPCR (Fig. 1C), 6 additional gene fusions with a lower number of reads were identified (Supplementary Table S7). This breast cancer case also harbored 204 nonsynonymous point mutations, including mutation of TP53 (G199E) and a known activating mutation of PIK3CA (H1047R; Supplementary Table S8). While this breast cancer case exhibited a number of amplifications and deletions (Supplementary Table S9), as expected (based on past clinical pathology data), this patient was negative for the ERBB2 amplification.
The fourth patient (MO_1081) with an FGFR2 fusion was a 57-year-old male with Gleason score 5+4 metastatic prostate cancer. Transcriptome sequencing of a brain metastasis revealed an interchromosomal fusion of SLC45A3 with FGFR2 in which the SLC45A3 noncoding exon 1 was fused to the intact coding region of FGFR2 (Fig. 1D and Supplementary Table S10). As SLC45A3 is a prostate-specific, androgen-regulated gene (20), the SLC45A3–FGFR2 fusion is predicted to drive overexpression of wild-type FGFR2. Importantly, FGFR2 exhibited outlier expression in the index case relative to our compendium of prostate cancer tissues (n = 84; Fig. 1D), and a similar rare case of FGFR2 outlier expression was identified in the Glinsky and colleagues (21) prostate cancer cohort (Supplementary Fig. S1A and B).
As we had identified novel FGFR2 gene fusions in cholangiocarcinoma, breast cancer, and prostate cancer, we next asked whether FGFR family fusions are present across carcinomas of different histologies. To address this, we analyzed RNA-seq data generated from an internal cohort of diverse tumors (n = 322) and The Cancer Genome Atlas (TCGA) effort (n = 2,053; Supplementary Table S11) for gene fusions using several bioinformatics approaches (see Methods). Including the initial 4 index cases, we identified 24 tumors or cell lines with FGFR1, 2, and 3 fusions (Fig. 2 and Supplementary Tables S12, S13, and S14). All of the gene fusions nominated expressed an FGFR family member as a 5′ or 3′ fusion partner with intact kinase domains suggesting potential actionability. 5′ FGFR fusions to BICC1, AFF3, CASP7, CCDC6, KIAA1967, OFD1, BAIAP2L1, and TACC3 (multiple exons) were identified and 3′ FGFR fusions to SLC45A3, BAG4, and ERLIN2 were identified. Cancer types harboring FGFR fusions were quite diverse and included cholangiocarcinoma (n = 2), breast cancer (n = 4), prostate cancer (n = 1), thyroid cancer (n = 1), lung squamous cell carcinoma (n = 6), bladder cancer (n = 5), oral cancer (n = 1), head and neck squamous cell carcinoma (n = 2), and glioblastoma (n = 2). FGFRs are known to exhibit tissue-specific splicing, resulting in IIIb and IIIc isoforms (22). Both IIIb and IIIc isoforms of FGFR2 and FGFR3 were evident in the RNA-seq data of the fusion cases, depending on cancer type (Supplementary Table S12).
As most of the diverse FGFR fusion partners contribute domains with known dimerization motifs, including coiled-coil, SAM, LisH, BAR, SPFH, and caspase (23–29), we hypothesized that oligomerization may serve as the common mechanism of activation of FGFR fusion proteins. Thus, we expressed selected epitope-tagged versions of the FGFR fusions in HEK 293T cells and looked for protein oligomerization by coimmunoprecipitation. For example, whereas FGFR3–BAIAP2L1, FGFR3–TACC3, FGFR2–BICC1, and FGFR2–CCDC6 interacted in vitro, wild-type FGFR2 and FGFR3 did not in the absence of FGF ligands (Fig. 3A and Supplementary Fig. S2). We also show that the isolated fusion domains provided by BAIAP2L1, TACC3, KIAA1967, CCDC6, and BICC1 interact in vitro as oligomerization domains (Supplementary Fig. S3), further supporting the notion of oligomerization-induced activation of FGFR kinase fusions. We additionally showed dimerization capability of the coiled-coil domain present in the FGFR2–CIT fusion identified recently in a lung adenocarcinoma by Seo and colleagues (ref. 30; Supplementary Fig. S3).
Unlike wild-type FGFR2 and FGFR3, overexpression of selected examples of FGFR fusions, including FGFR2–BICC1, FGFR3–BAIAP2L1, and FGFR3–TACC3, in 293T cells induced morphologic changes characterized by rounding up of cells (Supplementary Fig. S4). Overexpression of these FGFR fusion proteins also enhanced cell proliferation based on real-time cell imaging (Fig. 3B). To further show that FGFR fusion kinases are biologically active, we stably expressed FGFR fusions in benign immortalized TERT-HME cell lines. Stable lines harboring the FGFR3–BAIAP2L1, FGFR3–TACC3, and FGFR2–CCDC6 fusions showed expression of active FGFR fusion kinases (as shown by tyrosine phosphorylation of the fusion kinases) and enhanced proliferation of the cells (Fig. 3C–E). Activation of downstream mitogen-activated protein kinase ERK1/2 and the transcription factor STAT1 was also observed in the stable lines (Supplementary Fig. S5). In addition, the ERLIN2–FGFR1 fusion also produced an active FGFR kinase, as shown by tyrosine phosphorylation of the expressed fusion construct (Supplementary Fig. S6).
To evaluate the effects of pharmacologic inhibition of cells naturally harboring FGFR fusions, we assessed the sensitivity of bladder cancer cell lines to an FGFR small-molecule kinase inhibitor PD173074 (31). SW780 cells were characterized to have a fusion of FGFR3–BAIAP2L1 in this study and a study by Williams and colleagues (ref. 16; Supplementary Fig. S7A), whereas J82 and HT-1197 cells harbor activating point mutations of FGFR3 [K652E and S249C respectively (32), Catalog of Somatic Mutations in Cancer (COSMIC)]. Importantly, while the FGFR fusion-positive cell line SW780 was sensitive to nanomolar concentrations of PD173074, the FGFR3-mutant cell lines used here were not (Fig. 4A), suggesting that FGFR fusions may exhibit sensitivity to FGFR inhibitors, whereas some FGFR mutations are known to be resistant (33). Inhibition of proliferation was also shown with a second FGFR inhibitor, pazopanib, again showing sensitivity of the FGFR fusion-positive lines SW780 and RT4 (Supplementary Fig. S7B). PD173074 exerted a cell-cycle arrest effect on fusion-positive SW780 cells, but not on fusion-negative HT-1197 cells (Supplementary Fig. S8). Similar results for FGFR fusion-positive lines were obtained in vivo. SW780 xenografts exhibited decreased tumor growth with increasing doses of PD173074, whereas J82 xenografts did not (Fig. 4B). Expression of the FGFR3–BAIAP2L1 fusion in vitro induced ERK1/2 activation (Supplementary Fig. S5), and, similarly, fusion-positive SW780 xenografts exhibited strong ERK1/2 activation, which could be abolished by treatment with the FGFR inhibitor PD173074 (Fig. 4C). The RT4 urothelial carcinoma line harboring FGFR3-TACC3 fusion also exhibited sensitivity to FGFR inhibition in a xenograft model (Fig. 4B). Toxicity of PD173074 was monitored by assessment of mouse body weight (Supplementary Fig. S7C).
Further experiments using siRNA knockdown show the central role of FGFR3–BAIAP2L1 fusion in SW780 cell proliferation. Knockdowns using either FGFR3 of BAIAP2L1 siRNAs resulted in a dramatic reduction in cell proliferation in fusion-positive SW780 cells. In contrast, knockdown of FGFR3 or BAIAP2L1 did not have significant effects on cell proliferation in either fusion-negative cell line J82 or HT-1197 (Supplementary Fig. S9).
Sequencing and analysis of each of the 4 FGFR fusion-positive patients described in this study were carried out in a time frame of 5 to 7 weeks. The sequencing results were each presented at our bimonthly multidisciplinary precision tumor board for discussion and deliberation. The first patient with cholangiocarcinoma, MO_1036, who harbored the FGFR2–BICC1 fusion, underwent a conventional chemotherapy regimen in which her cancer progressed, chose not to pursue FGFR-directed therapy, and died 3 months after enrollment on this protocol. The second patient with cholangiocarcinoma, MO_1039, also harboring an FGFR2–BICC1 fusion, underwent conventional chemotherapy but did not show tumor shrinkage and was enrolled on an FGFR inhibitor clinical trial. The patient with metastatic breast cancer, MO_1051, harboring the FGFR2–AFF3 fusion, died of end-stage disease before the sequencing results were available. The patient with metastatic prostate cancer underwent irradiation of the brain (after brain metastasis resection) and continues to be maintained on hormonal treatment. Because of his brain metastasis, the prostate cancer patient was not eligible for an FGFR clinical trial.
Activating point mutations of FGFR1, FGFR2, FGFR3, or FGFR4 have been identified in a variety of cancers, including gliomas, bladder cancer, multiple myeloma, and rhabdomyosarcomas (34). Studies of hematologic diseases led to the identification of 3′ gene fusions of FGFR1 in myeloproliferative disorder (35) and 3′ FGFR3 fusions in peripheral T-cell lymphoma (36) and multiple myeloma (35). As described earlier, 5′ gene fusions of FGFR1 and FGFR3 with TACC1 and TACC3 have recently been identified in GBM in 2 studies (15, 37). Here, we identify potentially actionable 5′ and 3′ FGFR rearrangements across a diverse array of both common and rare solid tumors. Ten novel FGFR fusion partners were identified. In the Singh and colleagues (15) GBM study, the mechanism of activation of the FGFR fusions was proposed to be through mislocalization to mitotic spindle poles mediated by the coiled-coil domain of the TACC fusion partner. This presumably leads to mitotic and chromosomal segregation defects, triggering aneuploidy. In the Parker and colleagues (37) GBM study, increased expression through loss of the FGFR3 3′ UTR and miR-99a regulation was hypothesized as an activating mechanism. While these may be potential mechanisms in the specific case of the FGFR3–TACC3 fusion proteins in GBM, this likely does not explain the diverse array of fusion partners identified for FGFRs in this study. We propose a different, potentially more inclusive, model in which the FGFR fusion partners (e.g., BICC1, TACC3, CCDC6, BAIAP2L1, KIAA1967, CASP7, CIT, and OFD1) mediate oligomerization, which triggers activation of the respective FGFR kinase. Of note, we have not detected any FGFR fusions that result in simple truncation of the FGFR protein, despite prior investigations suggesting that 3′ truncating splicing isoforms encode activated FGFR2 proteins (38). The FGFR fusions detected have persistently exhibited substantial dimerization domain contributions from the 3′ fusion partner.
The SLC45A3–FGFR2 gene fusion identified in the index prostate cancer is quite interesting, as its pathogenic role is likely through a mechanism that is distinct from fusion protein oligomerization (shared by the other gene fusions tested). The entire open reading frame of FGFR2 is expressed under the control of an androgen-regulated promoter of SLC45A3, leading to the marked overexpression of FGFR2. The SLC45A3–FGFR2 fusion is analogous to the previously characterized TMPRSS2–ETS gene fusions characterized in more than 50% of prostate cancers (7). One would predict that this patient should respond to second-generation antiandrogens, such as MDV3100 (39), as well as FGFR inhibition. Another interesting observation in this study is the enhanced sensitivity to the FGFR inhibitor PD173074 of cell lines harboring an FGFR3 fusion relative to those that have an activating point mutation of FGFR3. While beyond the scope of this study, additional FGFR inhibitors and larger panels of FGFR fusions and FGFR-mutant cell lines will need to be studied to determine the broader applicability of these results. Clinical trials for several FGFR inhibitors are underway or in late-stage preclinical development (33, 40, 41). It will be important to enrich these early-stage clinical trials with patients harboring FGFR gene fusions, similar to the successful development of the small-molecule kinase inhibitor crizotonib in patients with lung cancer harboring the EML4–ALK gene fusion. The wide range of cancers in which FGFR rearrangements were detected in this study suggests that development of FGFR rearrangements is lineage-independent and emphasizes the importance of developing mutation-enriched clinical trials rather than trials based on tissue of origin. While each individual type of genetic aberration may occur at low frequency, the integrated sequencing approach identifies a wide range of informative genetic aberrations, potentially guiding the enrollment into numerous trials of diverse therapeutics.
In this study, we identified 4 patients with FGFR family gene fusions through an established clinical sequencing project called MI-ONCOSEQ. Combining these index patients with an analysis of transcriptome data from our internal tumor cohorts as well as the TCGA identified FGFR fusions in a wide array of cancers, including cholangiocarcinoma, GBM, squamous lung cancer, bladder cancer, breast cancer, thyroid cancer, oral cancer, head and neck squamous cell carcinoma, and prostate cancer. In addition to TACC1 and TACC3, we identified 10 additional FGFR fusion partners, as well as implicated 3 out of 4 FGFR family members (FGFR1, 2, and 3) in gene rearrangements. We also suggest a common mechanism of activation of these fusion proteins and show that FGFR gene fusion-positive cancers have enhanced susceptibility to FGFR inhibitors over activating point mutations of FGFR.
Clinical Study and Specimen Collection
Sequencing of clinical samples was conducted under IRB-approved studies at the University of Michigan (Ann Arbor, MI). Patients were enrolled and consented for integrative tumor sequencing, MI-ONCOSEQ (IRB# HUM00046018; ref. 3). Medically qualified patients 18 years or older with advanced or refractory cancer are eligible for the study. Informed consent details the risks of integrative sequencing and includes upfront genetic counseling. Biopsies were arranged for safely accessible tumor sites. Needle biopsies were snap frozen in optimum cutting temperature compound and a longitudinal section was cut. Hematoxylin and eosin-stained frozen sections were reviewed by pathologists to identify cores with highest tumor content. Remaining portions of each needle biopsy core were retained for nucleic acid extraction.
Cell Lines and Antibodies
Cell lines were purchased from the American Type Culture Collection and verified by next-generation transcriptome sequencing methods to identify known somatic mutations (COSMIC database). Oral cancer cell lines were obtained from their originating lab (A.-J. Cheng) and are not verified. Cells were grown in specified media supplemented with FBS and antibiotics (Invitrogen). Anti-c-MYC antibody was purchased from Sigma. Anti-V5 antibody was purchased from Life Technologies. Anti-FGFR3 antibodies were purchased from Epitomics and Cell Signaling. Antisera for phospho-FGFR, phospho-ERK1/2, pan-ERK1/2, phospho-STAT1, and pan-STAT1 were purchased from Cell Signaling. Anti-phosphotyrosine antibody clone 4G10 is from Millipore.
DNA/RNA Isolation and cDNA Synthesis
Genomic DNA from frozen needle biopsies and blood was isolated using the Qiagen DNeasy Blood & Tissue Kit, according to the manufacturer's instructions. Total RNA was extracted from frozen needle biopsies using the Qiazol reagent with disruption using a 5-mm bead on a TissueLyser II (Qiagen), and purified using a miRNeasy Kit (Qiagen) with DNase I digestion, according to the manufacturer's instructions. Total RNA was isolated from cancer cell lines using the TRIzol reagent (Life Technologies). RNA integrity was verified on an Agilent 2100 Bioanalyzer using RNA Nano reagents (Agilent Technologies). cDNA was synthesized from total RNA using SuperScript III (Invitrogen) and random primers (Invitrogen) for quantitative real-time PCR (qRT-PCR) analysis.
Preparation of Next-Generation Sequencing Libraries
Transcriptome libraries were prepared following Illumina's TruSeq RNA protocol, using 1–2 μg of total RNA. Poly(A)+ RNA was isolated using Sera-Mag oligo(dT) beads (Thermo Scientific) and fragmented with the Ambion Fragmentation Reagents kit (Ambion). cDNA synthesis, end-repair, A-base addition, and ligation of the Illumina indexed adapters were conducted according to Illumina's protocol. Libraries were size-selected for 250–300 bp cDNA fragments on a 3% Nusieve 3:1 (Lonza) agarose gel, recovered using QIAEX II gel extraction reagents (Qiagen), and PCR-amplified using Phusion DNA polymerase (New England Biolabs) for 14 PCR cycles. The amplified libraries were purified using AMPure XP beads. Library quality was measured on an Agilent 2100 Bioanalyzer for product size and concentration. Paired-end libraries were sequenced with the Illumina HiSeq 2000 (2 × 100 nucleotide read length). Reads that passed the chastity filter of Illumina BaseCall software were used for subsequent analysis.
Exome libraries of matched pairs of tumor/normal genomic DNAs were generated using the Illumina TruSeq DNA Sample Prep Kit, following the manufacturer's instructions. Three micrograms of each genomic DNA was sheared using a Covaris S2 to a peak target size of 250 bp. Fragmented DNA was concentrated using AMPure XP beads (Beckman Coulter), followed by end-repair, A-base addition, and ligation of the Illumina indexed adapters according to Illumina's protocol. The adapter-ligated libraries were electrophoresed on 3% Nusieve 3:1 (Lonza) agarose gels and fragments between 300 and 350 bp were recovered using QIAEX II gel extraction reagents (Qiagen). Recovered DNA was amplified using Illumina index primers for 8 cycles. The amplified libraries were purified using AMPure XP beads and the DNA concentration was determined using a Nanodrop spectrophotometer. One microgram of the libraries were hybridized to the Agilent SureSelect Human All Exon V4 at 65°C for 60 hours following the manufacturer's protocol (Agilent). The targeted exon fragments were captured on Dynal M-280 streptavidin beads (Invitrogen), and enriched by amplification with the Illumina index primers for 9 additional cycles. After purification of the PCR products with AMPure XP beads, the quality and quantity of the resulting exome libraries were analyzed using an Agilent 2100 Bioanalyzer and DNA 1000 reagents.
We used the publicly available software FastQC to assess sequencing quality. For each lane, we examined per-base quality scores across the length of the reads. Lanes were deemed passing if the per-base quality score boxplot indicated that more than 75% of the reads had >Q20 for bases 1–80. In addition to the raw sequence quality, we also assessed alignment quality using the Picard package. This allows monitoring of duplication rates and chimeric reads that may result from ligation artifacts and provides crucial statistics for interpreting the results of copy number and structural variant analysis.
Nomination of Gene Fusions
To identify gene fusions, paired-end transcriptome reads passing filter were mapped to the human reference genome and UCSC genes, allowing up to 2 mismatches, with Illumina ELAND software (Efficient Alignment of Nucleotide Databases) and Bowtie (42). Sequence alignments were subsequently processed to nominate gene fusions using the method described earlier (9). In brief, paired-end reads were processed to identify those that either contained or spanned a fusion junction. Encompassing paired reads refer to those in which each read aligns to an independent transcript, thereby encompassing the fusion junction. Spanning mate pairs refer to those in which one sequence read aligns to a gene and its paired-end spans the fusion junction. Both categories undergo a series of filtering steps to remove false positives before being merged together to generate the final chimera nominations. Reads supporting each fusion were realigned using BLAT (UCSC Genome Browser) to reconfirm the fusion breakpoint.
We annotated the resulting somatic mutations using RefSeq transcripts. HUGO gene names were used. The impact of coding nonsynonymous amino acid substitutions on the structure and function of a protein was assessed using Blocks Substitution Matrix scores. We also assessed whether the somatic variant was previously reported in dbSNP135 or COSMIC v5668.
Tumor content for each tumor exome library was estimated from the sequence data by fitting a binomial mixture model with 2 components to the set of most likely single-nucleotide variant (SNV) candidates on 2-copy genomic regions. The set of candidates used for estimation consisted of coding variants that (i) exhibited at least 3 variant fragments in the tumor sample, (ii) exhibited zero variant fragments in the matched benign sample with at least 16 fragments of coverage, (iii) were not present in dbSNP, (iv) were within a targeted exon or within 100 base pairs of a targeted exon, (v) were not in homopolymer runs of 4 or more bases, and (vi) exhibited no evidence of amplification or deletion. To filter out regions of possible amplification or deletion, we used exon coverage ratios to infer copy number changes, as described below. Resulting SNV candidates were not used for estimation of tumor content if the segmented log-ratio exceeded 0.2 in absolute value. Candidates on the Y chromosome were also eliminated because they were unlikely to exist in 2-copy genomic regions. Using this set of candidates, we fit a binomial mixture model with 2 components using the R package flexmix, version 2.3-8. One component consisted of SNV candidates with very low variant fractions, presumably resulting from recurrent sequencing errors and other artifacts. The other component, consisting of the likely set of true SNVs, was informative of tumor content in the tumor sample. Specifically, under the assumption that most or all of the observed SNV candidates in this component are heterozygous SNVs, we expect the estimated binomial proportion of this component to represent one-half of the proportion of tumor cells in the sample. Thus, the estimated binomial proportion as obtained from the mixture model was doubled to obtain an estimate of tumor content.
Copy number aberrations were quantified and reported for each gene as the segmented normalized log2-transformed exon coverage ratios between each tumor sample and matched normal sample (43). To account for observed associations between coverage ratios and variation in GC content across the genome, locally weighted scatterplot smoothing (LOWESS) normalization was used to correct per-exon coverage ratios before segmentation analysis. Specifically, mean GC percentage was computed for each targeted region, and a LOWESS curve was fit to the scatterplot of log2-coverage ratios vs. mean GC content across the targeted exome using the LOWESS function in R (version 2.13.1) with smoothing parameter f = 0.05.
Somatic point mutations were identified in the tumor exome sequence data using the matched normal exome data to eliminate germline polymorphisms. Parameters and computational methods were as previously described (44).
For RNA-seq gene expression analysis, transcriptome data was processed as previously described. Genes were nominated as exhibiting potential “outlier” expression relative to a cohort of n = 282 previously sequenced tissues using the following conditions: (i) the gene was required to have an expression value of at least 20 RPKM in the sample of interest; (ii) the gene was required to be at or above the 90th percentile relative to all previously sequenced tissues, of any type; (iii) the gene was required to have a fold change of at least 2 relative to the maximum reads per kilobase per million reads over all previously sequenced benign tissues; and (iv) the 25th percentile of the gene expression measurements over the previously sequenced tissues was required to be less than 50 RPKM. Collectively, these parameters target genes with (i) high absolute expression, (ii) high expression relative to previously sequenced tissues, (iii) high expression relative to all benign tissues, and (iv) expression that is not uniformly high across all tissues.
Partially redundant sequencing of areas of the genome affords the ability for cross-validation of findings. We cross-validated exome-based point mutation calls by manually examining the genomic and transcriptomic reads covering the mutation using the UCSC Genome Browser. Likewise, gene fusion calls from the transcriptome data can be further supported by structural variant detection in the genomic sequence data as well as copy number information derived from the genome and exome sequencing.
For validation of fusion transcripts, qRT-PCR assays were conducted. Total cDNAs of index cases and negative control samples were synthesized using SuperScript III System according to the manufacturer's instructions (Invitrogen). Quantitative RT-PCR was conducted using fusion-specific primers (Supplementary Table S15) with SYBR Green Master Mix (Applied Biosystems) on the StepOne Real-Time PCR System (Applied Biosystems). The PCR products were further analyzed by agarose gel electrophoresis. Relative mRNA levels of the fusion transcripts were normalized to the expression of the housekeeping gene GAPDH.
Inhibition of FGFR Receptors and Cell Proliferation Assay
Bladder cancer cells SW780, J82, and HT-1197 were seeded into 96-well plates in triplicate and allowed to attach before drug treatment. The FGFR inhibitor PD173074 (Selleck Chemicals) was added to the cultures at concentrations of 0, 5, 25, and 100 nmol/L. Relative cell numbers were measured by WST-1 assays at indicated time points following the manufacturer's instructions (Roche). To test the effects of the FGFR inhibitor pazopanib (Selleck Chemicals) on cell proliferation, SW780, RT4, J82, and HT-1197 cells were seeded into 24-well plates in quadruplicates and allowed to attach before drug treatment. Pazopanib was added to the cultures at concentrations of 0, 0.1, 0.5, and 1 μmol/L. Cell proliferation was determined by IncuCyte live-cell imaging system (Essen Biosciences).
Cloning and Expression of FGFR Fusions
The FGFR fusion alleles were PCR amplified from cDNA of the index cases or cell lines using the primers listed in Supplementary Table S15 and the Expand High Fidelity protocol (Roche). PCR products were digested with restriction endonuclease and ligated into the pcDNA3.1 vector (Invitrogen), which had been modified to contain a C-terminal MYC-epitope tag or V5-epitope tag. Expression constructs were transfected into HEK 293T cells using FuGene HD transfection reagent (Promega). Cells were harvested 24 hours after transfection for protein analysis. For stable line establishment in TERT-HME cells, FGFR fusion alleles were cloned into the pCDH510B lentiviral vector (System Biosciences), which had been modified to contain a C-terminal V5 epitope tag. Lentiviruses were produced with the ViraPower packaging mix (Invitrogen) in 293T cells using FuGene HD transfection reagent (Roche). Benign TERT-HME cells at 30% confluence were infected at a multiplicity of infection of 20 with the addition of polybrene at 8 mg/mL, and the cells were selected by 20 μg/mL puromycin. Stable pools of resistant cells were obtained and analyzed for expression of the FGFR fusion proteins by Western blot analysis with anti-V5 antibody. Cell proliferation was measured by IncuCyte imaging system as described above.
For the cell proliferation assay, HEK 293T cells were transfected with control vector or FGFR fusion constructs. Twenty-four hours after transfection, cells were trypsinized, resuspended in Dulbecco's Modified Eagle Medium (DMEM) containing 2% FBS, and plated in quadruplicate at 12,000 cells per well in 24-well plates. The plates were incubated at 37°C and 5% CO2 atmosphere using the IncuCyte live-cell imaging system (Essen Biosciences). Cell proliferation was assessed by kinetic imaging confluence measurements at 3-hour time intervals.
HEK 293T cells were grown to approximately 70% confluence in DMEM supplemented with 10% FBS, followed by transfection with MYC-tagged or V5-tagged expression construct alone or in combination using FuGene6 reagent (Promega). Twenty-four hours after transfection, cell pellets were lysed in lysis buffer (58 mmol/L Na2HPO4, 17 mmol/L NaH2PO4, 68 mmol/L NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, and protease inhibitors), followed by immunoprecipitation with tag epitope-specific antibodies (Sigma) and protein-G Dynabeads (Invitrogen). Precipitates were washed 3 times with IP Wash buffer (20 mmol/L Tris, pH 8, 2 mmol/L EDTA, 150 mmol/L NaCl, 1% Triton X100) and eluted in SDS-PAGE loading buffer at 95°C for 5 minutes. Immunoprecipitated proteins were separated on SDS-PAGE and detected by Western blotting with tag epitope-specific antibodies (Sigma).
siRNA Knockdown of FGFR3 and BAIAP2L1
SW780, J82, and HT-1197 bladder cancer cells were transfected twice with FGFR3-targeting siRNA, BAIAP2L1-targeting siRNA, or nontargeting siRNA (Thermo Scientific Dharmacon) using DharmaFECT1 reagent (Dharmacon). The siRNAs used were as follows: ON-TARGETplus FGFR3 L-003133-00-0005, ON-TARGETplus BAIAP2L1 L-018664-00-0005, and ON-TARGETplus Nontargeting pool. Twenty-four hours after transfection, cells were trypsinized and plated in triplicate at 8,000 cells per well in 24-well plates. The plates were incubated at 37°C with 5% CO2 atmosphere in the IncuCyte live-cell imaging system (Essen Biosciences). Cell proliferation rate was assessed by kinetic imaging confluence measurements at 3-hour time intervals.
Mouse Xenograft Models
Five week-old male C.B17/SCID mice were procured from a breeding colony at University of Michigan, maintained by Dr. Kenneth Pienta. Mice were anesthetized using a cocktail of xylazine (80 mg/kg, intraperitoneal) and ketamine (10 mg/kg, intraperitoneal) for chemical restraint. Bladder cancer cells SW780 (2 million cells for each implantation site) or J82 (5 million cells for each implantation site) were resuspended in 100 μL of 1× PBS with 20% Matrigel (BD Biosciences) and were implanted subcutaneously into flank region on both sides. Eight mice were included in each experimental group. All tumors were staged for 2 weeks (SW780 cells) and 3 weeks (J82 cells) before starting the drug treatment. Xenografted mice with palpable tumors were treated with a FGFR inhibitor PD173074 (Selleck Chemicals) dissolved in 5% ethanol in corn oil (intraperitoneal). Mice in control group received 5% ethanol in corn oil as vehicle control. Tumor growth was recorded weekly by using digital calipers, and tumor volumes were calculated using the formula (π/6) (L × W2), where L = length of tumor and W = width. Any decrease in the body weight of mice was monitored biweekly during the course of the study. All experimental procedures involving mice were approved by the University Committee on Use and Care of Animals at the University of Michigan and conform to their relevant regulatory standards. Tumor tissues from xenografted SW780 cells were harvested and lysed in radioimmunoprecipitation assay buffer containing protease/phosphatase inhibitors for Western blot analysis.
Disclosure of Potential Conflicts of Interest
A.M. Chinnaiyan is a consultant to Life Technologies, co-founder of Compendia Biosciences, which is now owned by Life Technologies, and advisor to Ventana/Roche and Gen-Probe/Hologic.
Conception and design: Y.-M. Wu, F. Su, X. Cao, S. Roychowdhury, M. Talpaz, D.R. Robinson, A.M. Chinnaiyan
Development of methodology: Y.-M. Wu, F. Su, S. Kalyana-Sundaram, B. Ateeq, A.-J. Cheng, D.R. Rhodes, D.R. Robinson, A.M. Chinnaiyan
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Su, X. Cao, J. Siddiqui, S.A. Tomlins, S. Roychowdhury, M. Hussain, F.Y. Feng, M.M. Zalupski, M. Talpaz, K.J. Pienta
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-M. Wu, F. Su, S. Kalyana-Sundaram, N. Khazanov, B. Ateeq, R.J. Lonigro, P. Vats, L.P. Kunju, S.A. Tomlins, P. Wyngaard, S. Sadis, S. Roychowdhury, K.J. Pienta, D.R. Rhodes, D.R. Robinson
Writing, review, and/or revision of the manuscript: Y.-M. Wu, F. Su, R. Wang, L.P. Kunju, J. Siddiqui, M. Hussain, F.Y. Feng, M.M. Zalupski, M. Talpaz, D.R. Robinson, A.M. Chinnaiyan
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Su, X. Cao, R. Wang, S.-F. Lin, A.-J. Cheng, J. Siddiqui, S. Roychowdhury
Study supervision: Y.-M. Wu, S. Roychowdhury, D.R. Rhodes, D.R. Robinson, A.M. Chinnaiyan
This project is supported in part by the Prostate Cancer Foundation, the NCI Early Detection Research Network (U01 CA111275), NIH RO1CA132874-01, a Department of Defense Era of Hope Scholar Award, and the University of Michigan Comprehensive Cancer Center Prostate SPORE (P50CA 069568). A.M. Chinnaiyan is supported by a Stand Up To Cancer-Prostate Cancer Foundation Prostate Dream Team Translational Cancer Research Grant. Stand Up To Cancer is a program of the Entertainment Industry Foundation administered by the American Association for Cancer Research (SU2C-AACR-DT0712). A.M. Chinnaiyan is also supported by the Alfred A. Taubman Institute, the American Cancer Society, the Howard Hughes Medical Institute, and a Doris Duke Charitable Foundation Clinical Scientist Award.
The authors thank Terrance Barrette and Doug Gibbs for hardware and database management, Shanda Birkeland, Mary Pierce-Burlingame, and Karen Giles for assistance with sample and manuscript preparation, Rachell Stender for animal care, and Dan Hayes for referral of patients to the MI-ONCOSEQ program. The authors also thank the larger MI-ONCOSEQ team, including clinical research coordinator Lynda Hodges, Amy Gursky for patient sample procurement, cancer geneticists S. Gruber and J. Innis, bioethicists J. Scott Roberts and Scott Y. Kim, genetic counselors J. Everett, J. Long, and V. Raymond, and radiologists E. Higgins, E. Caoili, and R. Dunnick.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).
- Received February 5, 2013.
- Revision received April 3, 2013.
- Accepted April 3, 2013.
- ©2013 American Association for Cancer Research.